Fault early warning algorithm based on classifying and clustering

Abstract

The invention discloses a fault early warning algorithm based on classifying and clustering, comprising the steps of S1, performing supervised anomaly detection, to be specific, using a classification model to train website data to obtain faulty data and non-faulty data; S2, performing non-supervised anomaly detection, to be specific, clustering the faulty data into multiple datasets, and analyzing and detecting faults; S3, performing half-supervised anomaly detection, to be specific, using part of labeled samples of high confidence to process information of the rest labeled samples; S4, processing unbalanced datasets, to be specific, using a sampling method to balance the datasets, and highlighting features of abnormal data to finish fault early warning. The classifying and clustering methods are used herein to perform mining, internal connections of data objects at abnormal points are fully considered to intend to train a feature model of abnormal data, a mining method of unbalanced data is introduced, features of faulty data are highlighted, and good classifying and clustering are achieved. The algorithm of the invention is simple and efficient.

Description

technical field [0001] The invention relates to the technical field of fault early warning algorithms, in particular to a fault early warning algorithm based on classification and clustering. Background technique [0002] Website failure data is website-oriented data, which includes text information (user questions and feedback information), website link information (scene id), access records (PV, UV, etc.); website failure event data is relatively rare, But this does not mean that they are absolutely irregular. A data object is different from other data objects (that is, anomalies), possibly because it belongs to a different type or class; Anomaly sources that we have not considered, and there may be multiple anomaly sources in the data set, and their underlying causes are often unknown. Fault warning technology is transparent to the causes of these anomaly sources, and is committed to discovering objects that are significantly different from other objects . [0003] Most...

AI Technical Summary

Problems solved by technology

[0003] Most of the existing fault warning technologies are for abnormal point detection, but the abnormal point detection does not consider the connection between abnormal points, and detects the abnormal point data as outliers or mutation points, and considers data objects with low density and significant changes It is an abnormal object. This type of algorithm does not require prior statistical data model training. The mining method is relatively simple and rough. Although the efficiency is high, the early warning effect is not satisfactory.